NL8420060A - DATA COMPRESSION SYSTEM AND METHOD FOR PROCESSING DIGITAL SAMPLED SIGNALS. - Google Patents

Description

8 ^ L u υ τυ υ N.O. 33361

Data compression system and method for processing digital sampled signals.

RELATED APPLICATIONS

This is a "continuation-in-part" application of U.S. Patent Application Serial No. 202,457 filed October 31, 1980 by Charles S. Weaver, entitled Method and Apparatus for Digital Data 5 Compression.

BACKGROUND OF THE INVENTION

Systems equipped with means for converting analog signals to digital form, after which compression filtering and Huffman encoding of the signals are used for recording or transmission to a remote location, together with playback or receiving means, which include a Huffman decoder, a digital reconstruct! filter and means for returning the decoded and filtered signals to the analog form are known from the aforementioned U.S. patent application, serial number 202,457 and from an article by U.E. Ruttimann and H.V. Pipberger, entitled "Compression of the ECG by Prediction or Interpolation and Entropy Encoding", IEE Transactions on Biomedical Engineering, Vol. BME-26, No. 11, biz.

613-623, November 1979. A similar system is known from an article by K.L. Ripley and J.R. Cox Jr., entitled "A Computer System for 20 Capturing Transient Electrocardiographic Data", Pro. Comput. Cardiol., Biz. 439-445, 1976. The present invention limits the average bit rate of an analog to digitally converted audio signal, such as a music signal, an electrocardiogram or electroencephalogram gram signal, to allow its digital transmission over a simple transmission line and / or to enable recording and playback of a significant amount of signals using a relatively small amount of a recording medium using known digital recording and playback techniques.

SUMMARY OF THE INVENTION

Audio signals, such as music, which are broadcast or recorded are converted into digital form by an analog digital converter. The digital signals are then fed to a digital compression filter to form digital compressed signals. The compressed audio signals are supplied to an encoder, such as a truncated Huffman encoder, to encode these signals. The digital outputs of this encoder are recorded by means of a digital recording device and / or broadcast to a remote receiver station. In a playback device or a receiving station, the encoded signal is decoded by a decoder, after which the decoded signal is supplied to a digital decompression! efil ter. The output of the decompression filter is converted into analog form by means of a digital-analog converter to provide a reproduction of the audio signals. A combination of a compression and a decompression filter is used such that the average bit length of the recorded or broadcast digital signal words is minimal. The transfer function of the digital compression filter has zero points on the unit circle in the Z plane at an angle of zero degrees; the transfer function of the digital decompress! efilter has poles on or within the unit circle in the Z-plane at an angle of zero degrees. The compression filtering operation is performed without truncation or rounding, while the decompression filtering operation is performed with truncation or rounding. Outside the zero degrees angular positions of the zeros of the compression filter transfer function, zeros on the unit circle in the Z plane may also be present at ± 41.41 °, ± 60 °, ± 90 °, + 120 ° and / or 180 °. A decompression filter tuned thereto is used; this filter has poles at or within the unit circle near the location of the zero points of the compression filter. The resulting frequency characteristic of the entire system is high-pass for the analog signals and may include one or more low-frequency notches. For music signals, the resulting filter has a low cutoff frequency between 0-15 Hz to pass 25 audio frequency signals in an interval from about 15 Hz to 20,000 Hz.

An unstable combination of compression and decompression filter is obtained when the transfer function of the decompression filter has poles on the unit circle in the Z plane. Such an embodiment requires the transmission of the output signals from the

Huffman encoder using an error check code and error detecting means to detect errors in the transmission to the Huffman decoder. As a result of detecting an error in such a digital signal transmission, an error signal at -35 is generated, which error signal is applied to the decompression filter to momentarily move the poles of the filter more inwardly within the unit circle, allowing, that the system recovers from the supplied error signals.

The above-described error signal detection and inward movement 40 of the poles of the transfer function in the Z-plane of the decompression 8 A 2 0 0 6 0 3 siter due to the error detection can be applied in those systems, which have a stable combination of compression - and have a decompression filter to accelerate recovery from the error signals.

Instead of an error check code and error signal detecting means in those systems, which have an unstable combination of compression and decompression filter, the system may operate by periodically feeding a series of actual signal values from the A / D converter to the the reconstruction filter, whereby the reconstruction filter is started periodically when errors occur. In the case of compression of music signals, a transmission of actual signal values every 6 to 16 milliseconds can be convenient.

The number of consecutive actual values required to be transmitted periodically depends on the order of the decompression filter; the number of actual signal values supplied is equal to this order.

BRIEF DESCRIPTION OF THE DRAWINGS

The invention will be better understood from the following description with reference to the accompanying drawings. In the drawings, where like parts from different angles 20 are indicated by like reference numerals, shows:

Fig. 1A together with FIG. 1B a block diagram of a data reduction system; a digital recording and transmission portion is shown in Figure 1A and a playback and reception portion is shown in Figure 1B;

Fig. 2 is a waveform and graphical representation of signals such as those occurring at the various locations in the data compression system depicted in FIGS. 1A and 1B;

Fig. 2A shows a frequency characteristic of a high-frequency attenuation and a high-frequency amplification filter, which are arranged at the input and the output of the data reduction system, respectively; FIG. 3 is a graphical representation of coded difference signals showing the format used which has been used to encode those difference signals outside the predetermined signal interval;

Fig. 4 an image showing the relationship between the probability that a digitally sampled signal value will occur at a certain quantization level and the magnitude of the quantization level;

Fig. 5 the zeros of the transfer function of a second order compression filter on the unit circle in the Z plane; Fig. 40 6 a number of z-transformed zeros which can be used in the compression fliter included in the present invention;

Fig. 7 is a view showing the frequency characteristic of three different zero-compression flashes on the unit circle of the z-transformed in some of the positions shown in FIG. 6;

Fig. 8 is a block diagram showing details of a compression filter of the type that can be used in the present invention; FIG. 9 is a table showing a truncated Huffman code of the type that can be used in the present invention;

Fig. 10 the zero-pole pattern of a combination of compression and reconstruction fliter, which can be used in the present system and which results in a stable system without the need for error detection;

Fig. 11 shows the frequency characteristic of the combination of compression and reconstruction filter with the zero-point poles pattern shown in FIG. 10;

Fig. 12A with 12B show a similar block diagram as FIG. 1A with FIG.

1B; here, however, the system includes a control bit generator and error detecting means for moving the poles of the reconstruction filter inward when a bit error is detected;

Fig. 13 is a zero point pole pattern of a combination of compression and reconstruction filter, with poles of the reconstruction filter momentarily moved inward to facilitate recovery of the transitions;

Fig. 14 is a block diagram showing details of a reconstruction filter that can be used in the present invention; FIG. 15 a zero point pool pattern of another combination of compression and reconstruction filter, which can be used in systems incorporating the present invention; and shows

Fig. 16 is a graphical representation of Huffman encoded signals contained in a system in which the actual digital signal values are periodically transferred to the reconstruction filter, the latter filter being periodically started up.

First of all, reference is made to Fig. 1A, which shows the digital recording and transmission portion of a data compression system and which includes an analog-to-digital converter (A / D converter) to convert an analog audio signal f (t) into digital form, where 8420060 the nth sample of the analog-to-digital converter 20 is designated fn. An analog signal 22 is indicated under A in Fig. 2, which serves as an input signal for the analog-digital converter 20. By way of example, it should be noted that the audio input signal may consist of a music signal, which is located in a frequency interval from about 15 to 20,000 Hz. The shape of the output signal of the analog-to-digital converter is shown under B in Fig. 2 and contains samples fn_x to fn + i of words of equal length. The analog-to-digital converter 20 operates at a sampling rate, which is determined by 10 control signals from the timing and control unit 24, which control signals are applied over the timing line 26. As used herein, the line 26 of the timing and control unit 24 represents a number of output lines from the time control unit, one or more of which are supplied to the system elements for correct timing and control of the system. Input signals are also supplied to the timing and control unit over line 28 to control them with signals from various other system elements. The A / D converter 20 operates in the usual manner at a fixed sampling rate and with a fixed word length output. For the sake of description, and certainly not as a limitation, it should be noted that the A / D converter can operate at a sampling rate of 44 KHz and with a 14-bit word length.

The output of the A / D converter 20 is supplied to a digital compression filter 30 through a digital filter 23 which attenuates the high frequency portion of the digital audio frequency signal from the A / D converter 20 to reduce signal entropy. The frequency characteristic of the filter 23, along with the frequency characteristic of the digital filter 75, is included in the playback and receive portion of the system shown in Fig. 2A.

For the sake of simplicity, the digital output signal of the filter 23 as well as the digital input signal is indicated by fn. It will be understood that an analog filter with a similar frequency characteristic can be included for the input of the A / D converter 20 instead of the digital filter 23 at its output.

For the purposes described here, the digital compression filter as shown includes an estimation element 32 and a subtractor 34. The estimation element 32 provides an estimate of fn, denoted here by f ^, which estimate is based on the actual samples occurring both before and after the sample to estimate fn.

a

Estimators for providing such an estimated value fn 8420060 6 are of course known. A difference signal Δπ is output from the compression filter 30; this difference signal consists of the difference between the actual input signal fn and the estimated signal value fn, obtained by subtracting the estimated value from the actual 5 value in the subtractor 34 according to: Δη = ^ n "^ n (!)

The differential signal and Δπ, Δη + 1> Δη + 2 »..... Δη + ι are shown in the graphical representation of the output signal of the compression filter under C in fig. In accordance with a feature of the present invention, the arithmetic operations of the digital compression filter 30 are performed without truncation or rounding, while the arithmetic operations of the associated digital decompression or reconstruction filter as described below are performed with truncation or rounding. As shown in Fig. 2 under C, the output of the compression filter includes uncut compressed signals, which are 18 bits long.

It will be understood here that the present invention is not limited to the use of the illustrated compression filter in which the output Δη is formed by the difference between the actual input

a

output signal fn and an estimated value fn. Other compression flashes can be used with a different transmission, in which the output Δη of the compression filter is not a direct function of the difference between the actual input fn and an estimated value thereof fn. The use of the term "difference" signal values Δη intends to also indicate the output of other suitable compression flashes.

The compressed signal values Δη are supplied via a switch 35 to an encoder 40, wherein a truncated Huffman code is used to encode the supplied signals. The Huffman encoding is known from U.S. Patent Application Serial No. 207,728 filed November 17, 1980 entitled "Method and Apparatus for digital Huffman Encoding" by Charles S. Weaver, which application is assigned to the same assignee as the present invention. Everything known from this U.S. Patent Application Serial No. 207,728 is specifically incorporated herein by reference. Briefly, the Huffman encoding technique takes advantage of the fact that the compression filter reduces entropy of the output signal Δη so that a reduction of the total number of bits in the Huffman encoded signal-to-input signal can occur. A single codeword is assigned to infrequently occurring difference signals and supplied as a label for the actual difference signal value Δη. In Fig. 1A, the output of encoder 40 is indicated by h (An) and under D in Fig. 2, the values h (An), h (An + i) etc. represent coded values of Δπ, Δη + ι, etc. The most frequently occurring value of Δη (here zero) is encoded using the shortest codeword. A truncated Huffman code is known from United States patent application serial number 207,728, which code 10 can be easily incorporated by using a simple encoder and decoder. The output of encoder 40 includes codewords for the most frequently occurring values of Δπ, together with a combined codeword label and actual value of the compressed signal Δη for less frequently occurring values of Δη. For the sake of clarity, it should be noted that when the compressed signal value exceeds ± 3, the actual compressed signal Δη is output along with a codeword label at the encoder output. From Fig. 3, which depicts different encoded compressed values, it will be apparent that the encoded value for Δη + 2 includes a label along with the actual compressed signal Δη + 2, where Δη + 2 is an infrequent represents compressed signal value; i.e. a value outside the interval of ± 3.

The encoded signals from encoder 40 are recorded and / or broadcast to a remote receiver. For recording purposes, the encoder output is connected through a switch 48 to a recording device 50 to record the encoded difference signals as labeled h (An) signals. With the switch 48 in the other broken line position, the encoder output is supplied to a buffer memory 52 and from there to a digital modem 54 for transmission over the transmission line 56. In certain embodiments of the In the present invention, control bits are generated to be recorded and / or broadcast with the encoded compressed signals h (4n).

In some embodiments of the invention, digital input signals fn are sometimes supplied to the input of the Huffman encoder through a switch 35, which signals serve to start or restart the cooperating digital reconstruction filter as described below.

40 Recorded encoded digital signals, such as those recorded with the recording device 50 in Figure 1A, are reproduced using the system distributed in Figure 1B, which system includes a playback unit 60. The recorded encoded digital signals from the playback unit 60 are supplied via a switch 564 to a decoder 66 for the truncated by a truncated

Huffman code to decode coded signals. In the decoder 66, the Huffman encoded words are converted into the original compressed signals Δη. Where a Huffman encoded word represents a labeled actual compressed signal, the tag is taken from it and the actual compressed signal without the tag is applied to the decoder output. The encoders and decoders that can be used in the present invention are described in detail in the aforementioned U.S. Patent Application Serial No. 207,728. Encoding and decoding are described in more detail below under the heading "Encoding-decoding".

The compressed signals Δπ from decoder 66 are supplied to a reconstruction or decompression filter 70 through a buffer memory 72. The decoder outputs are provided at a slightly varying rate, while the buffer memory 72 is inserted to easily satisfy the input speed requirements of the reconstruction filter 70. The reconstructive! 1 ter 70 converts the compressed signals Δπ into sample signals fn (off) of equal length, which are very close to the input sample signals fn of the compression filter 30. As already noted above, a feature of the present invention is that compression filtering without chipping and decompression filtering with chipping or clipping is used. In Fig. 2 under F, the aborted reconstruction filter output signals fn (off), fn + i (off) etc. are depicted and include 24-bit words. Without breaking, the reconstruction filter would process the words to a length of about 36 to 40 bits, which is not available for consumer products at a reasonable cost. In the following it will become clear that a suitable data compression with minimal distortion is obtained by using a combination of a compression and a decompression filter, with no interruption in the compression filtering but an interruption in the decompression filtering.

A digital-to-analog converter (A / D converter) 74 converts the signal samples fn (off) from the digital reconstruction filter 70 into analog form to reproduce the analog signals. A digital filter 75, 40 which amplifies the high frequency components of the output signal is included in the connection of the output of the digital reconstructive! 1b to the D / A converter. The frequency characteristic of the filter 75 is shown in FIG. 2A along with the frequency characteristic of the input filter 23. For the sake of simplicity, the same symbol fn (off) is used at the input and output of the filter 75. It will be understood that an analog filter with a similar frequency characteristic can be connected to the output of the D / A converter instead of the digital filter 75. The receiver's timing and control unit 76 provides the timing signals 10. for the various elements of the receiver over line 78 for appropriate timing for the operation of the receiver.

Also, control signals from the various elements of the receiver over the line 80 are applied to the unit 76 to control them.

In the case of transmission without recording, the encoded signals are transmitted over line 56 (from FIG. 1A to FIG. 1B) to a digital modem 82 in the receiver. The modem output is stored in the buffer memory 84, while the buffer memory output is applied through the switch 64 in the broken line position to the decoder 66 for decoding and subsequent processing in the manner described above. QUANTIZATION OF ANALOGUE MUSIC SIGNALS

For the sake of clarity and not limitation, an analog music signal is considered as an input signal of the system of the present invention described herein.

The entropy of a binary x bit converted from analog to digital (A / D) sample is

2X

H (q) - ΣΖ - Pi 1092 ^ 1 (2) 30 i = 1 where 2X possible values of the sample are present and ΡΊ · represents the probability that the i-th possible value will occur. When the magnitude of the quantization level is indicated by q35 and, for the sake of simplicity, it is assumed that the i-th quantization giving the i-th value ranges from q (i-1) to qi. Then, as shown in Fig. 4, the analog-to-digital conversion signal will fall in the interval q (i-1) to qi. In Figure 4, the darkened surface represents the probability that the signal f (t) falls in the i-th quantization.

8420060 10

Furthermore, it is assumed that the magnitude of the quantization level q is small compared to the standard deviation cr of the analog music signal. As the word length of the A / D converter increases by one bit, the size of the quantization is divided in half and 5, as indicated by dashed lines in FIG. 4, two quantization bins are formed from the original bin. For small values of q, the area on either side of the vertical dashed line will be approximately equal, with the probability that f (t) will fall in either of the two new bins is approximately equal to P-j / 2. Hence 10 the contribution of the two new bins to the entropy of the n + 1 bit long word is approximately equal to 15 ^ [t1 log2 Y1] = .2 [Pi (log2Pi-l)] = - log · log2Pi + Pi (3) 20

The entropy of the (x + 1) bit long word is 25 h (-H = ^ - pl '° 92pi + J2 pi - H (q) + 1 (4)

From the foregoing, it appears that as the bit length increases, the increase in entropy will converge to one bit for each bit added to the word length. When the first quantization bin is centered around zero (the usual case), the argument is somewhat more complex; however, the result is the same.

The Pi and the entropies have been evaluated by numerical integration with different ratios of <y to q for the Gaussian distribution. The following Table 1 of calculated entroopin increments at different ratios of y y / q shows that the entropies increase very closely but one bit each time q is divided in half or when the word length increases by one.

8420060 11

Table 1

CALCULATED INCREASE IN THE ENTROPY WHEN REDUCING THE SIZE OF THE QUANTIZATION

Entropy σ q (bits) 0.5 0.58 1 1.16 2 1.93 4 2.82 8 3.77 16 4.75 32 5.73

COMPRESSION FILTERING OF QUANTIZED MUSIC SIGNALS

The average word length of a Huffman encoder, such as the encoder 40, to which the output of the compression filter 30 is applied is limited as follows: 5 H (q) <average word length <H (q) + 1 (5)

If a coefficient in the equation (s) used to realize the compression filter has a non-integer value, then the quantization level at the filter output will be limited; i.e. the minimum difference between possible output values will decrease, while H (q) will increase. For example, consider the following two compression equations to realize the compression filter transfer: 15 4n = fn - 2fn-l + fn-2 (6) and 20 Δ „= f„ - 2fn_! + 2-® + lf „_i + f„ -2 - 2-i »+ lf„ .2 + 2-2fflfn-2 (7) where m is a positive integer.

The Z transformed of equation (6) has two zeros at (1.0) and equation (7) has two zeros at (1-2m, 0) 8420060 12 in the Z plane. The Δη in equation (7) will have values with a mutual difference of 2-2fflq. When m is large, the value ö * of the two filters will be approximately the same, but the ratio of σ / quantization level will differ by a factor of 2-2m. Hence, the entropy of equation (7) will be about 2m more bits than the entropy of equation (6), and the average bit length after Huffman encoding will be about 2m bits longer.

It should be noted that multiplying the right side of equation (7) by a factor 22m reduces the quantization level from 10 to q, but the standard deviation then increases by a factor 22m so that the ratio remains unchanged.

Filter weights against word! Narrow

The general form of a compression filter difference equation is 15

a

Δη = Σ2 aifn-i + l (8) i = l 20 where a-j is a constant. If a-j can be represented by a binary number of finite length, then a-j can be expressed by 25 a, - - ΣΙ at 2J '(9) j where at * 0 or +1 and j can have positive or negative values. A negative value of j and a value of not equal to zero mean that fn_i is shifted to the right and increased; j0 bits must be added to the least important end of the arithmetic word, where j0 is the most negative value of j at a value of b-jj that is not zero.

The Z transformed of equation (6) is 35 G (Z) = (1 - 2z-l + z-2) = (1 - zl) 2 (10) which can be indicated by two zeros in (1.0) in the Z-plane as shown in fig. 5. The frequency characteristic at f0 of 40 a digital filter with only zero points (for example equation (6) or equation (7)) is the product of the distances from the point exp 0420060 13 ( j2ufT) to each of the zeros and the gain constant (equal to 1 in the equation (10)), where f represents the frequency and T the time between samples. The frequency characteristic of equation (10) is therefore: 5 R (f0) - d2 (11)

When there are n zeros in (1.0), 10 R (f0) = dn (12)

These compression flashes reduce entropy for the following reason: if the A / D sampling rate is 44 x 10 samples per second, then the point (-1.0) on the unit circle corresponds to a frequency of 22 KHz. The centers of music spectra will usually be less than 1 KHz, so that most spectral points correspond to points on the unit circle that are near (1,0). The value of d (and dn) will be much less than one and the integral of the spectrum time value dn as a function of Θ ((where Θ = 2rf) will be 20 less than the variance of the input spectrum (K ^ l) variance means reduced entropy.

The value of d is greater than one when θ> 60 ° (f = 7.33 KHz), so spectral components above 7.33 KHz are amplified with dn on filters with all zero points in (1.0). There is a value of n such that with an increase of n above this value, the total energy above 7.33 KHz is amplified more than the total energy is weakened below 7.33 KHz. This value of n minimizes the output variance and entropy because the input and output q are equal when K = 1. This is evidenced by the following: 30 G (Z) = K (1 - zl) n 35 - '£ Htf & TT * < 13> n = K J2. 9ιζ "ί i = 0 40 where a ·, · are the constants used in equation (8) and K is a gain constant. The aj are integers that can be developed (as in equation (9)) without negative 8420060 14 value of j, so that q is not reduced.

It should be noted that different values of K do not change the ratio <r / q or entropy when K is a power of two because the input word is only shifted. The value of n that minimizes the entropy for K = 1 also minimizes it for other values of K while the minimum entropy is obtained when K is a power of two.

Two other zero positions on the unit circle that do not decrease q are in (-1.0) and (the complex pair) in (0.1) and (0, -1). The Z-ge-10 transformed are: G (Z) = (1 + z'1) 11 (14) [n zeros in (-1.0)] and 15 G (Z) = (1 - z-2) n (15) [n zeros in (0,1) and n zeros in (0, -1)].

The other two complex pairs of positions, which do not change q, have the following transformed: (G (Z) = (1 - z-1 + z-2) n (16) (making n zeros on the unit circle at angles of + 60 ° are placed with respect to the origin and n zeros below -60 °) and G (Z) = (1 + z-1 + z'2) n (17) (so n zeros are placed below + 120 ° and n zeros below 30 -120 °) The above are the only zero positions inside or on the unit circle that do not decrease q There are none outside the unit circle resulting in a satisfactory reconstruction filter.

Another zero point position of importance for music data compression is the complex pair at angles of ± 41.41 ° on the unit-35 circle. The transformed is G (Z) = 1 - 1,5z "l + z-2 (18)

This angle corresponds to 5.06 KHz while q is divided by 2 for 40 for each complex pair.

8420060 15

In accordance with an aspect of the present invention, a compression filter is used, the Z-transformed thereof having zeros in (1.0) and on the unit circle and at least one of the above complex pair positions (these are + 41.41 °, ± 60 °, 5 ± 90 °, ± 120 ° and 180 °). The above zero point positions are depicted in Fig. 6. As mentioned before, these zero point positions on the unit circle minimize entropy, while the combination of zero point positions used depends on the spectrum of the signal to be compressed. For example, zeros can be placed 10 below ± 60 ° to reduce the portion of the output variance that results from high frequencies (from about 3 to 14 KHz) so that more zeros can be applied in 1.0. Fig. 7 shows the frequency characteristic of three different compression filters, which filters zero points on the unit circle of the Z transformed at 15 °; 0 ° and ± 60 °; and have 0 °, ± 90 ° and ± 120 °. It will be found that the useful zero point position for entropy minimization allows for a compression filter design with a wide range of frequency characteristics.

Of course, there are limitations when the number of zero-20 increases. The amount of calculations required is directly proportional to the number of zeros, while the arithmetic filter word length increases by at least one for each additional zeros. Also, the recovery during bit error reconstruction will take more time if the number of zeros has increased.

25 After selecting the compression filter transfer function, the entropy that will be obtained can be estimated in the following way: the music spectrum S (f) is measured and the integral ƒ | G (Z) | 2 Sz (ej2irfT) dz (19 ) 30 is integrated along the unit circle from (1.0) to (-1.0), where

Sz (eJ'2irfT) = S (f).

35

The square root of the integral is the <r of the compression filter output. Table 1 can now be used to estimate H (q). Compression filter

While it will be appreciated that standard digital techniques 40 can be used to accomplish the previously described 8420060 16 compression filter transfer functions, including the use of a programmed digital calculator, a block diagram of a second order digital compression filter suitable for realizing the equation (6) shown in Fig. 8, to which Fig. 5 is now referred. The illustrated compression filter includes a series of shift registers 102, 104 and 106 in which the successive sample signals from the A / D converter are passed through the filter 23. In Fig. 8, for the purpose of description, registers 102, 104 and 106 are shown with samples fn, fn_i and fn-2, respectively. For 14 bit samples, 14 bit registers are used. The register outputs are connected to a digital multiplexer 108 for selectively connecting the sample signals to an arithmetic and logic unit (ALU) 110. The multiplexer 108 and ALU 110 are controlled by the timing and control unit 24.

As noted above in the description of Fig. 1A, the digital compression filter 30 may include an estimation element 32, of which

a

the output is formed by an estimated sample value fn based on the actual sample values fn_i and fn + i, which appear before and after the sample fn to be estimated. Frequently, streams-20 use the estimation elements that provide an output signal% = alfn + 1 + a2fn-l (20), choosing the coefficients ai and a2 to minimize the mean square error of the difference Δη, with Δη = fn - f “n, as indicated in the aforementioned equation (1). For ai = a2 = 1, equations (1) and (20) can be combined to 30 Δπ = fn + i - 2fn + fn_i (21) (Note that equations (6) and (21) are equivalent).

The equation (21) can be used by the illustrated compression filter to generate the compressed signal δπ.

An estimated value of the sample fn is obtained using the samples located on either side of fn, these are the samples fn_i and fn + i, but not fn itself. Under control of the unit 24, the words fn_i and fn + i are introduced into the ALU 110 40 via the multiplexer 108 and added there. The actual sample 8420060 17 star fn is then fed to the ALU 110 through the multiplexer 108 and multiplied by 2. Multiplying by 2 simply involves a shift from the bits to the most significant bit. It with 2 steps

a

the multiplied actual sample is subtracted from fn to provide the compressed signal value Δη at the output of the ALU 110, which signal value is then applied to the encoder 40. The calculation in the ALU 110 is performed with a word which is long enough to avoid truncation or rounding errors. It will be understood that data compression in the apparatus 10 described herein involves estimating a sample value by interpolation.

HUFFMAN CODING AND DECODING

As noted above, Huffman encoders and decoders suitable for use in the present system of encoding and decoding the compression filter output are known from copending U.S. Patent Application Serial No. 207,728 filed November 17, 1980, entitled "Method and Apparatus for Digital Huffman Encoding in the name of the inventor of this application, which application is incorporated herein by reference.

Reference is made to Fig. 9, which illustrates an example of an aborted Huffman code for illustration and not limitation. Indicated is a table of compressed signals Δη ranging from ± 3 together with a codeword for these signals, the length of the codeword and the relative probability of occurrence of these compressed signals. The most common compressed signals Δη (here between ± 3) are assigned a code word. The probability that Δη has a value to which a codeword is assigned is high, eg 0.98. Code words of varying length are assigned to these compressed signals, with the most frequently compressed signal being assigned the shortest code word. In the table, the most frequently occurring compressed signal, Δπ = 0, is assigned the shortest code word and the least frequently occurring compressed signal, Δη = -3, is assigned the longest code word. All other compressed signals outside the ± 3 interval are designated "elsewhere" in the table; these signals are assigned a codeword, which, as described above with reference to Figs. 2 and 3, contains a label for the actual compressed signal value Δπ, which is then transferred to the Huffman decoder by recording, transmission over a connecting pipe or the like. The 40 system is clearly not limited to use with the indicated broken Huffman code. Extra compressed signals Δη can be assigned a codeword, while other codewords can also be used.

RECONSTRUCTION FILTERING

5 Entropy against reconstruction filter stability

For an accurate reconstruction of the digital music signal supplied to the digital compression filter 30, the transfer function of the reconstruction filter 70 must be the inverse of the transfer function of the compression filter 30. (Two other necessary conditions for accurate reconstruction are that there are no overshoot or undershoot errors in the filter calculation operation and the word length of the compression filter output is not truncated.)

As indicated above, the minimum entropy is obtained when the zeros of the compression filter transfer function are on the unit circle. The exact inverse has poles on the unit circle in the same positions as the zero points of the compression filter. Such a reconstruction filter is unstable. Such instability is sufficient until a bit error occurs, after which "initial conditions" erroneously and randomly cause the reconstruction filter to diverge to saturation. Two different systems are described below for use in a device, in which the compression reconstruction filter combination is unstable, which systems correct the effects of bit errors.

Briefly, a system includes periodically transferring some actual digital signal values fn to the digital reconstruction filter 70 to periodically restart it. This, of course, requires a blocking of the signal and without overly complex and very fast logic, the loss of data from the point where the error occurs to the end of the block.

Another system involves using check bits and error detecting means to generate a bit error signal whenever an error is detected. The error signal is used to momentarily move the poles of the compression filter 70 inward into the unit circle; in this short time, the filter recovers from the consequences of errors without the need to restart the filter with actual signal values fn. By placing the poles of the reconstruction filter within the unit circle, the filter becomes stable and will attenuate erroneous "initial conditions" resulting from errors. Under these conditions, the filter is stable and 8420060 19 no blocks are required for error recovery. Stable filter combinations of this type are also described in more detail below.

Compression filter zero points within unit circle 5 insufficient

It should be noted here that the poles of an exact inverse are not on the unit circle if the zero points of the compression filter are not on the unit circle. Such a compression reconstruction filter combination, as will become clear from the following example, is not practical for music data compression. If the transfer function of the compression filter has two real zeros near the point (1,0), these must be at a distance from the unit circle, which is less than π x 200/22000 (the distance around the unit circle corresponds to 200 Hz) . A greater distance means that the low frequency components will not be attenuated equally. Since 0.00909π k 2 "?, the zero point will be determined by t1 - (1 - 2-7) z-1].

Hence, if two zeros were used, a coefficient of 2 ~ 14 would be present and 14 bits would have to be added at the least important end of the filtering calculation. Without interruption, about 14 bits should be added to the entropy and only small data compression will be possible.

Another compression reconstruction filter combination, which is not suitable for use in the present system, also includes a structure in which the two zero points of the compression filter and the poles of the reconstruction filter are away from the unit circle. With this construction, there is no arithmetic termination of the word length, but the compression filter output is interrupted at a length one or two bits longer than the word length of the A / D converter. If the output quantization level is equal to q0, then the quantization noise power will have a variance of 35 t - # <22> and the noise will be white. The reconstruction distortion will be equal to 40 the output noise caused by a noise generator at the input of the reconstruction filter, which has a variance determined by equation (22).

Since the input noise samples are white and statistically independent, it can be shown that the output noise variance is σο2 - £ Si2 (23) i = o or 10 / \ \ 1/2 ff0 - * q0 (5Z 9i2) (24) 15 where gj is the i-represents the value (at the i-th sample time) of the impulse response of the reconstruction filter. In other words, the square root of the sum of the squares of the impulse response samples is the standard deviation multiplier. This multiplier is calculated for different pool positions by solving the respective difference equation. It has been determined from calculations that the noise power induced by such degradation of the compression filter output signal is too large or concentrated in such a small portion of the signal bandwidth that an unwanted tone occurs in the music output signals. . Thus, breaking off the compression filter output words is unsatisfactory for music data compression.

Reconstructive! It is not an exact inversion of the compression filter

If there are no bit errors in the transfer of the output from the compression filter 30 to the input of the digital reconstruction filter and there is no breakdown of the output from the compression filter 30, then the output of the Huffman decoder 66 is identical to the output of the compression filter 30. It will therefore be appreciated that the transfer from the input of the compression filter 30 to the output of the reconstructor 70 is simply the product of the transfer functions of the two filters 30 and 70. Another data compression system employing the present invention includes a compression reconstruction filter combination in which the zero points of the compression filter are located at certain positions on the unit circle to reduce the entro-40 pie and the corresponding poles of the reconstruction filter are located within the unit circle near said zero point 8420060 21 for the sake of convenience stability. The frequency response and stability of such compression reconstruction filter combinations can be easily calculated. Consider, for example, a compression reconstruction filter combination, in which the compression filter has two 5 zeros in (1.0), while the reconstructive 1ter has two poles in (1-0.00195.0). The poles zero point pattern of such a compression filter in cascade circuit with a reconstruction filter is shown in Fig. 10, while the frequency response of the filter combination is shown in Fig. 11. As can be seen from Fig. 11, the combination produces a very smooth high-pass filter with a cut-off frequency of 18 Hz. With this filter combination, a bit error recovery is possible within 20-30 ms. It should be noted here that the reconstruction filter 70 used herein preferably includes a digital calculator programmed for the desired reconstruction filter operation.

Reflections on the word! Narrow in the reconstruction filter

A stable reconstruction filter that works without truncation will require a large arithmetic length. For example, the 18 Hz real-pole filter described above requires at least 34 bits for the calculation operation (0.00195 = 2 "^), at least 9 bits per pole at the least important end, and 1 bit per pole at the most important end. word length of the A / D converter is 14 bits, thus a 4-pole configuration will require a much longer arithmetic word length Currently, calculators working with such a long word length are not practical for consumer music data compression.

Fortunately, the reconstruction filter arithmetic can be broken down to practical length under negligible system degradation. The arithmetic cutoff noise is analyzed in much the same way as the quantization analysis. For this analysis, noise generators with a noise power q2 / 12 (a generator for each coefficient) are added to the filter input and the input-output standard deviation multiplier is calculated. The value of q is the quantization level of the abbreviated arithmetic.

35 For an 18 Hz two-pole reconstruction filter, the multiplier is 3227. The arithmetic word length must then be 12 bits longer than the word length of the A / D converter, or the arithmetic cross-cap will be greater than the A / D. quantization noise power. A 14 bit A / D conversion requires a 26 or 27 bit reconstruction filter calculation 40. If the compression filter has two real zeros in (1.0) and 8420060 22 two complex poles at 7 KHz on the unit circle, then a reconstruct! filter are used with a complex pair of poles in the 20 Hz Butterworth position and a complex pair at 7.33 KHz along the unit circle and 100 Hz away from the unit circle. The standard deviation factor of such a reconstruction filter is 32, or 5 bits, while the frequency response of the compression reconstruction filter combination is substantially the same as in Figure 11, except for a narrow indentation at 7 KHz. With a 24-bit calculation operation, hardly any quality deterioration of the signal occurs. For example, digital calculators with a 24-bit reconstruction filtering operation are available at an acceptable cost for a system according to the invention.

SYSTEM WITH ERROR DETECTION AND MOVABLE RECONSTRUCT! EFILTER POLES

Figures 12A and 12B show a modified embodiment of the invention, generating control bits for recording and / or transmission along with the encoded digital compressed signals. In the playback device and / or the receiver unit, an error signal is generated on the basis of the errors detected by means of the control bits, which is used to transmit the poles of the digital reconstruction filter briefly inwards or further inwards within the unit circle in the unit circle. move z-plane without changing the pole angle. For an unstable reconstruction filter, momentary inward movement of the poles from the unit circle results in a stable filter combination, correcting playback and / or transmission errors without the need to restart the filter. For a stable reconstruction filter, momentary inward movement of the poles from the unit circle means accelerated recovery of erroneous signals.

First, refer to Fig. 12A, which illustrates the digital recording 30 and transmission portion of a modified embodiment of a data compression system, which modified form includes the use of control bits. The system shown in Fig. 12A is similar to that in Fig. 1A and includes an analog-to-digital converter 20, a digital compression filter 30, a Huffman encoder 40, a switch 48, a recording device 50, a buffer memory 52 a modem 54 and a timing and control unit 24, all of which may be of the same type as shown in FIG. 1A and described above. It should be noted that an analog high-frequency attenuation filter 23A is applied to the input of the A / D converter, which filter has the same function as the digital filter 23 in Fig. 1A.

8420060 23

In the embodiment of the invention shown in Figure 12A, a control bit generator 90 is present in the connection of the Huffman encoded signal h (Δη) to the recording device 50 or the modem 54, depending on the position of the switch 48. The 90 control bits generated by the control bit-generator 90 are added to the stream of Huffman encoded signals for recording and / or transmission along with said encoded digital compressed signals. Numerical schemes for generating control bits and for error detection using such control bits are well known and do not require detailed description. It should be noted here that recording devices and modems often include control bit generator means for generating control bits to be added with the data stream to be recorded or broadcast.

The recorded encoded digital signals along with control 15 bits as recorded using the recording apparatus 50 are reproduced using the playback unit 60 shown in FIG. 12B, to which reference is now made. The signals emitted by modem 54 (FIG. 12A) are transferred via line 56 to modem 82 (FIG. 12B). The switch 64 connects the output of the playback device or the output of the modem to an error checking circuit 92, where the signal flow is checked for bit errors. When an error is detected, a bit error signal is generated, which signal is applied over line 94 and through switch 96 to the digital reconstruction filter to momentarily slide in the poles 25 of the filter.

Control bit signals are taken from the signals of the playback device 60 and / or the modem 82 by the error checking means 92, and the Huffman encoded digital compressed signal stream h (An) from the error detector is supplied to the Huffman decoder 66, which decoder of the type can be as described in Fig. 1B above. From the Huffman decoder, the digital compressed signals Δη are supplied through the buffer memory 72 to the digital reconstruction filter 70A. Like the reconstruction filter 70 in Fig. 1B, the reconstruction filter 70A operates with abort and converts the compressed input signal Δη into sample signals fn (out) of equal length, which sample signals very close the input sample signals fn of the compression filter 30 (Fig. 12A) to approach. A digital analog converter 74 converts the signal samples fn (off) into analog form f (t). An analog high-frequency gain filter 75A is present at the output of the D / A converter, which filter has the same function as the filter 75 in Fig. 1B; namely, restoring the amplitude of the high-frequency signals attenuated in the filter 23A.

Poles within the unit circle in the Z plane 5 As noted above, in one embodiment of the present invention, a compression reconstruction filter combination is applied, where the zero points of the compression filter are located at certain points on the unit circle to reduce entropy, while the reconstruction filter has corresponding poles within the unit circle near said zero points for stable operation. The recovery of the bit filter reconstruction filter is accelerated by momentarily moving the poles of the reconstruction filter inward from the unit circle in the Z plane whenever an error signal is detected in the error checking circuit 92.

In Fig. 13, the zeros and poles of the transfer function of a compression reconstruction filter combination are depicted. The zero points of the compression filter are indicated on the unit circle below zero degrees, while a pair of reconstruction filter poles are depicted near the zero points and normally at a distance of 0.00195 within the unity circle. It should be noted that this combination of zero points and poles is the same as that described in Fig. 10 above. In the system depicted in Figure 13, however, the reconstruction filter poles are momentarily moved inward after receiving a bit error signal from the error checking circuit 92 through line 94. For the sake of clarification 25, the poles are shown after moving to a 0.0625 point within the unit circle to obtain a rapid recovery from the error. After a short time, for example 50 ms, the poles of the reconstruct turn! efil ter back to their normal position, which is the point 0.00195 within the unit circle below zero degrees.

30 Difference equations for a two pole reconstruction under 0 ° within the unit circle are: yn = 2Δη + ayn-i = 2δπ + yn-l -2-rayn-l (25) 35 ^ n = -Vn + a ^ nl = -Vn + ^ n-1 “2” m ^ nl (26) 40 where: a = l-2 “m and m = an integer.

A reconstruction filter satisfying equations (25) and 8420060 (26) is shown in Figure 14, to which Figure will now be further referenced. The reproduced reconstruction filter 70A includes a 4 to 1 digital multiplexer 130 with an input 132 to which compressed signals Δη are supplied from the decoder 66. The output 5 of the multiplexer 130 is supplied to an arithmetic and logic unit, ALU, 134, where the required multiplications by shifting, adding and subtracting take place under the control of the time control and control unit 76A.

The output of the ALU 134 is connected to the input of a 1 10 to 2 digital demultiplexer 138. An output of the demultiplexer 138 is connected to a register belonging to a pair of shift registers 140 and 142 connected in series via a line 144. The other demultiplexer output is connected via a line 146 to a single shift register 148. The value of yn, determined in the ALU, is recorded in the register 140, while the previous value of yn is shifted from the register 140 to the register 142. The third register 148 is supplied with the sample value fn (off), as calculated in ALU 134.

The output signals of registers 140, 142 and 148 form the input signals for ALU 134 via multiplexer 130. In use, the value stored in register 148 represents the signal fn-i (off). It can be seen from equation (25) that the value yn is calculated using the input signals Δη, 25 and yn-1 respectively supplied to the ALU 134 over the line 132. Equation (26) shows that the sample value fn (off) is calculated using the input signals yn, respectively, from registers 140 and 148, fn-1 (off)

As long as m equals a finite integer, the reconstruction filter 70A operates stably and no filter start-up or restart is required. If no bit errors are present, the filter works with a relatively large value of m, for example m = 9, to place the poles of the filter to the side of the unit circle at 0.00195 from the unit circle. When an error is detected by the error checking circuit 92, a smaller value of m is used, for example, m = 4, moving the poles of the filter inward to a point 0.0625 away from the unit circle. The bit error signal from the error checking circuit 92 (Fig. 12B), which is applied over line 94 to ALU 134, controls the value of m 40 used in the implementation of equations (25) and (26) 8420060 26 by simply controlling the number of places to be scrolled to perform the indicated multiplication by the factor 2 "ra. When an error is detected, the contents of a corresponding ALU register are not shifted so far to the right -5 for a nominal time value (for example 50 ms) when performing the multiplications by 2_m, moving the reconstruction filter poles inward from the unit circle to accelerate a transfer recovery After this short period of time, normal operation returns of the reconstruction filter with poles close to the 10 unit circle.

Poles normally on the unit circle in the Z-plane

As noted above, another embodiment of the present invention involves a compression reconstruction filter combination, where the zero points of the compression filter are located in certain positions on the unit circle to reduce entropy, and the reconstruction filter has corresponding poles, which also extend throughout the normal operations on the unit circle are in the same positions as the zero points; that is, as long as bit errors are omitted during operation. However, when a bit error is detected by the error checking circuit 92, the poles of the reconstruction filter 70A are momentarily moved within the unit circle to obtain stable reconstructive filtering and error recovery. This embodiment can be realized by using the above-described receiver unit or the playback device shown in Fig. 12B and the reconstruction filter 70A shown in Fig. 14. However, in the absence of transients, the reconstruction filter 70A operates with poles on the unit circle, as shown in FIG.

15. In Fig. 15, two zeros of the compression filter 30 are shown, arranged on the unit circle in the Z plane below zero degrees, 30 while during operation without bit errors occurring, the two poles of the reconstruction filter 70A are located in the same position on the unit circle, at a point where m is infinite.

If a bit error signal from the error checking circuit 92 is present, the reconstruction filter poles are momentarily moved inward of the unit circle at zero degrees. For the sake of clarification, m, as shown, has been changed to value 4. Under these conditions, the reconstruction filter quickly recovers from the error without the need to restart or restart the filter by transmitting actual signal values fn thereto. Here, it should be noted that upon starting the operation, the reconstruction filter poles 8420060 27 were briefly moved within the unit circle to prevent generation of any saw band function at its output.

The invention is clearly not limited to moving the reconstruction filter poles inward to a single location in the presence of an error signal. Different values of m can be used, with the filtering operation stepped through several different pole locations during bit error recovery. For example, values of m equal to 2, 4, and 7 can be used, 10 during operation switching to m is 2, then to m is 4, and finally to m is 7 before returning to the original value of m on or within the unit circle in the Z-plane.

PERIODIC TRANSFER SYSTEM OF fn 15 Another embodiment of the present invention utilizes a compression reconstruction filter combination, wherein the zero points of the compression filter are located at specific points on the unit circle in the Z plane to reduce entropy, and the reconstruct ! efil ter has corresponding poles on the unit circle in the same places as the zero points, which poles are fixed and not moved inward. As noted above, such a compression-reconstruction filter combination is unstable and any transients result in a random output from the reconstruct! efil ter. In order to minimize the effects of such transients, the reconstruction filter is periodically restarted during operation by adding a number of actual signal values fn. The apparatus shown in Figures 1A and 1B can be used for such an operation.

The signal flow transferred for this operation is shown in Figure 16, to which Figure now refers. In addition to the Huffman-encoded difference signal and h (An + 2) ..... h (4n + j) are

Huffman encoded signal values h (fn), h (fn + i), etc. are periodically transmitted by moving the switch 35 periodically to the position shown in the broken line shown in Fig. 1A. With the switch 35 in the broken line position, a series of actual signal values fn is periodically supplied to the Huffman encoder 40 to be encoded and then recorded or broadcast. The number of consecutive signal values fn emitted is equal to the order of the reconstruction filter.

40 In the signal stream shown in Fig. 16, two 8420060 28 consecutive signal values fn are periodically encoded to reconstruct a second order! periodically restart the filter 70. The coded signals h (fn) etc. are depicted with a label and the actual signal value fn. The label used differs from the "elsewhere" label, used to indicate those signals which are outside the predetermined interval of compressed signal values Δη. The tag portion of the coded signals h (fn) is indicated in tag 16 with tag # 2 to distinguish it from the "elsewhere" tag.

The required number of encoded signal values h (fn) is periodically transmitted, for example every 10 milliseconds, as shown in FIG. 16 to periodically restart the associated digital reconstruction filter 70. By periodically restarting the reconstruct! filter, it is not necessary to operate the reconstruction filter with poles within the unit circle in the Z plane, since any 15 sawtooth output caused by transient signals is eliminated within 0 to 10 milliseconds.

The invention has been described here in detail in accordance with the requirements of the "Patent Statutes" with various other changes and modifications self-explanatory to those skilled in the art. For example, many of the functions described can be realized by using a digital calculator with suitable calculation routines. It is intended that these and other such changes and modifications will fall within the spirit and scope of the invention as defined in the appended claims.

8420060

Claims (26)

1. A data compression system for processing fixed length digital sampled signals, comprising a digital compression filter, which provides compressed signals responsive to the digital sampled signals, the transfer function of the digital compression filter having zero points on the unit circle in the Z plane at substantially zero degrees from the origin, which digital compression filter has a gain, which is a power of two, and performs arithmetic operations without truncation or rounding, a digital encoder responsive to the output of the digital compression filter, and applies a truncated Huffman code, which digital encoder encodes only those compressed signals located in a predetermined signal interval, and which labels those compressed signals located outside this predetermined interval, a digital decoder, means to convert the output signal of the digital c transfer to the digital decoder to decode it, and a digital reconstructor which responds to the output of the digital decoder for reconstructing its output signal, the digital reconstruction filter having the transfer function poles at or within the unit circle in the Z-plane at substantially zero degrees from the origin, which digital reconstruction filter performs arithmetic operations with truncation to obtain truncation errors in its output. 2. Data compression system according to claim 1, characterized in that the transfer function of the digital compression filter also has zeros on the unit circle in the Z plane at angles from the origin of at least one of the following pairs of angle values: ± 41.41 ° , ± 60 °, ± 90 °, ± 120 ° and ± 180 °, the transfer function of the digital reconstruction filter poles 35 on or within the unit circle in the Z-plane with angular positions essentially corresponding to the zero points of the transfer function of the digital compression filter. Data compression system according to claim 1, characterized in that the transfer means use an error check code and 8420060 error detecting means for detecting errors in the transfer from the output of the digital encoder to the digital decoder and to generate a error signal when an error in the transmission is detected, 5 means are present which respond to an error signal from the error detecting means to momentarily move the poles of the digital reconstructive filter within the unit circle in the Z plane, without the pole being angled changed to facilitate recovery of the digital reconstruction filter from the detected errors. Data compression system according to claim 3, characterized in that the transfer function of the digital reconstruction filter has poles on the unit circle in the Z plane, which poles are briefly moved within the unit circle in response to an error signal from the error detecting means. Data compression system according to claim 3, characterized in that the transfer function of the digital reconstruction filter has poles within the unit circle in the Z plane, which poles are briefly moved further within the unit circle in response to an error signal from the error detecting means. . 6. Data compression system according to claim 1, characterized in that the transfer function of the digital reconstruction filter has poles on the unit circle in the Z-plane at angular positions, which essentially correspond to the zeros of the transfer function of the digital compression filter, and in that means are provided to periodically supply a number of consecutive digital sampled signals to the digital reconstruction filter to periodically restart operation thereof. 7. Data compression system according to claim 6, characterized in that the digital sampled signals supplied to the digital reconstruction filter are supplied thereto via the digital encoder, the digital decoder and the transfer means. 8. A data compression system according to claim 7, characterized in that the digital encoder is periodically operable to periodically label a number of consecutive digital sampled signals to transmit them to the digital reconstruction filter via the transfer means and the digital decoder. Data compression system according to claim 8, characterized in that the digital encoder operates every 6 to 16 milliseconds to periodically supply successive digital sampled signals 40 to the digital reconstruction filter. 8420060 10. Data compression system according to claim 1, characterized in that the compressed signals of the digital compression filter are related to the difference between the sampled input signal thereof and an estimated value thereof, which estimated digitally sampled signal is obtained by using of sampled signals on either side of the digital sampled signal to be estimated. 11. Data compression system according to claim 1, characterized in that an analog-to-digital converter is present, from which digital sampled signals are obtained by analog-to-digital conversion of analog signals, and a digital-to-analog converter for converting the digital output signals of the digital reconstruction filter into analog form. Data compression system according to claim 11, characterized in that the analog signals comprise music signals. Data compression system according to claim 12, characterized in that the digital sampled signals are obtained from the analog-to-digital converter at a speed between 30 and 50 KHz. Data compression system according to claim 1, characterized in that the transmission means comprise means for recording the coded signal from the digital encoder and means for playing back the signal recorded by the recording means. Data compression system according to claim 1, characterized in that the transfer means comprise a first and second modem 25 and a transmission line for connecting the modems. 16. Data compression system according to claim 1, characterized in that an input filter is present for attenuating high frequencies in the input signal for the digital compression filter, as well as an output filter for amplifying high frequencies in the output signal of the digital reconstruction filter. Data compression system according to claim 16, characterized in that the input and output filters are digital. 18. Digital data compression system, comprising a source for fixed length digital sampled data signals, a compression filter for generating compressed signals, a reconstruction filter, which responds to compressed signals from the compression filter, reproducing the digital data signals, and means to transfer the output of the compression filter to the input of the reconstruction filter, and a method of processing the 40 compression and reconstruction filter to obtain a reduction in signal transducer with low signal distortion, and the improvement including the operation of the digital compression filter 1) with a transfer function with zeros on the unit circle in the Z plane at least at an angle of 0 ° measured from the origin, 2) with a gain that is a power of two, and 3) without truncation of the signal, and processing of the reconstruction filter 1) with a transfer function with poles on or within the unit circle in the Z-plane at the same angular positions measured from the origin as the zero points of the digital compression filter, and 2) with arithmetic truncation of the word length to obtain truncation errors in the output signal of the digital reconstruction filter. The method as defined in claim 18, wherein the digital compression filter operates a transfer function with additional zero points on the unit circle under at least one of the following 15 pairs of angular positions measured from the origin: ± 41.41 °, ± 60 °, ± 90 °, ± 120 ° and ± 180 °, and where the digital reconstruction filter operates with an additional poles transfer function at or within the unit circle in the Z plane at the same additional angular positions as the zero points of the digital compression filter. The method as defined in claim 18, wherein the poles of the reconstruction filter are momentarily moved inward from the unit circle in the Z plane due to transient errors in the transfer of the output signal from the compression filter to the input of the reconstruction filter in order to accelerate the correction of the errors in the operation of the reconstruction filter. 21. The method as defined in claim 18, wherein the digital reconstruction filter operates with poles on the unit circle in the Z plane and periodically applies a number of consecutive digital sampled signals to the reconstruction filter to periodically restart its operation. while the number of consecutive digital sampled signals supplied to the reconstruction filter is equal to the order of the reconstruction filter. 22. Data compression system for processing a stream of 35 digital sampled signals, comprising a digital compression filter, which provides compressed signals responsive to the digital sampled signals, the transfer function of the digital compression filter having zeros on the unit circle in the Z plane below an angle of 0 ° and below at least 40 one of the following pairs of angle values, measured from the origin: 8420060 ± 41.41 °, ± 60 °, ± 90 °, ± 120 ° and ± 180 °, a digital reconstruction filter with a transfer function with poles on or within the unit circle in the Z plane at substantially the same angular positions as the zeros of the digital compression filter, 5 and means for transferring the output signal of the digital compression filter to the digital reconstruction filter. Data compression system according to claim 22, characterized in that. that the transfer means includes a Huffman encoder 10 for truncated Huffman encoding of the output of the digital compression filter and means responsive to the encoder output. Data compression system according to claim 22, characterized in that the transfer means comprise a bit control generator 15 and means for generating an error signal when a transient error is present, and means for briefly moving the error signal controlled by this error signal. poles of the reconstruction filter inwardly within the unit circle without changing the angular positions thereby obtaining accelerated recovery of the detected transient errors. 25. Data compression system for processing a stream of digital sampled signals, comprising a digital compression filter, which provides compressed signals responsive to the digital sampled signals, the transfer function of the digital compression filter having zeros on the unit circle in the Z plane , a digital reconstruction filter with a transfer function with poles on or within the unit circle in the Z plane at essentially the same angular positions as the zeros of the digital compression filter, means for transferring the output signal of the digital compression filter to the digital reconstruction filter, and means responsive to transient errors in the transmission of the output of the digital compression filter to the digital reconstruction filter momentarily move the poles of the reconstruction filter within the unit circle in the Z-plane without the angular positions of the poles can be changed essentially in order to facilitate the correction of the transient errors. +++++++ 8420060 AMENDED CONCLUSIONS [only claim 1 modified (1 sheet)] 1. Data compression system for processing fixed length digital sampled signals, comprising a digital compression filter, which provides compressed signals responsive to the digital sampled signals, the transfer function of the digital compression filter having zeros on the unit circle in the Z plane at essentially zero degrees from the origin, which digital compression filter has a gain, which is a power of 10 two and performs arithmetic operations without truncation or rounding, a digital encoder responsive to the output of the digital compression filter and a variable code word length implements, a digital decoder, means to transfer the output of the digital encoder to the digital decoder to decode it, and a digital reconstruction filter responsive to the output of the digital decoder for reconstructing filtering Exit s signal thereof, of which digital reconstruction filter the transfer function has poles on or within the unit circle in the Z-plane at substantially zero degrees from the origin, which digital reconstruct! efil performs arithmetic operations in which truncation occurs to obtain truncation errors in its output signal. +++++++ 8420060